Will super-smart artificial intelligence be the ultimate cheat code for beating the market? Probably not … but its could ensure you retire with 25% more than you otherwise would.
In this TechFirst we chat with the CEO of Qraft USA, Robert Nestor. Qraft is an AI company focused on investing. The company has a billion USD “assets under AI” in Korea and is expanding to the US. While the AI is active in crafting the strategy and execution, humans still make the final decisions.
Support TechFirst: Become a $SMRT stakeholder
The result, so far, is about a 1% improvement on what financial advisors would otherwise get. While that doesn’t sound huge, that’s significant over your earning and investing lifespan.
Watch our chat here, or keep scrolling for the full audio or transcript:
AI for investing: will AI investment engines take over from human financial advisors?
Transcript: Qraft CEO Robert Nestor on AI investment engines
TF241 – Qraft
(This transcript has been lightly edited for length and clarity.)
John Koetsier: Is AI the ultimate cheat code when it comes to investing? Investing money is notoriously hard. Studies of trading have shown that few traders actually outpace the market, which is one reason why index funds that invest in a wide basket of companies tend to outperform the vast majority of mutual funds. But can AI make a difference? And if so, is there hope for the rest of us?
To chat about it, we’re meeting with Robert Nestor, who is the CEO of Qraft Technologies. They’re building an AI-enabled financial technology. Welcome to TechFirst, Robert.
Robert Nestor: Thanks, John. Thanks for having me.
John: Hey, super pumped to have you. And yeah, as I was saying, as we were chatting earlier, anybody who’s got some money invested in the stock market somewhere is pretty interested in this, right? I mean, how are decisions made about where your money is being allocated?
Let’s start here: What does Qraft do?
Robert: So Qraft is at the intersection of technology and asset management. And specifically, we’re using AI processes to augment and accentuate human intuition. That is what artificial intelligence is all about, and I think, John, that’s largely missed on a lot of people. They think it’s just about programming lines and let the robot go. It is not exactly that. So we’re at that intersection. We were founded six years ago in Korea, largely Asian-PAC based originally, and now we’re on the ground in the U.S.
And it is our job and our vision to transform investment decision-making with artificial intelligence, and we’ll talk more about what that means.
John: So, that’s cool. There’s still a role for humans. That’s good on the one hand.
On the other hand, everybody just wants, like, where’s my magic eight-ball? Where’s my cornucopia? Where do I feed my seed $1,000 in, and boom, it goes to the moon, right?
You know, nothing to be done there. Maybe talk a little bit about how it works, how the technology works, and how the interplay works with people.
Robert: Yeah. So I alluded to it a moment ago. I think what’s often lost on people is what the intention is here. The intention very directly is to teach machines through artificial intelligence, machine learning — some of these labels we’ve all heard — machine learning, deep learning, and natural language processing.
But the point is to teach machines the power of human intuition but to deliver it at scale. So, you know, the human mind is an amazing thing, incredible, thoughtful decision-making, and nothing can replace the intuition of a human mind.
But the limitations of a human mind are largely two. One is that intuition cannot be applied easily at scale. And that is what artificial intelligence is fundamentally about. But it’s also the human mind and the human being lets bias creep in at times, emotional bias, and the like. And the role of artificial intelligence is not to completely remove that intuition but to actually harness it and deliver it at scale with techniques that have only recently been available.
So the human being is never completely removed but what we’re doing is taking human intuition and scaling it.
John: Well, it’s good to know that there’s a human in there. On the other hand, of course, intuition, hey, I’m gonna buy the dip and it keeps dipping, right? [laughing] So that does come into it.
This comes out of obviously two kind of areas. Your name is Qraft. There’s sort of quant investing, which has been popular and in demand for probably a decade now or something like that. And then this whole world of machine learning and masses of data and applying it. How does this differ from pure quant investing?
Robert: So, we get that question all the time. Quant’s been in … quantitative investment’s been around for decades. The fundamental difference in the application of artificial intelligence and investment decision-making and traditional quant is, traditional quant still started with a hypothesis driven by a human.
You know, in the simplest of terms, if stocks have these specific characteristics, they should outperform. And a thoughtful quant goes about either proving or disproving, often disproving, that hypothesis but it still originates with the human being.
What artificial intelligence does, it doesn’t start with any hypothesis. It just looks to take in and analyze massive data sets at a scale as, you know, I said a few moments ago, unavailable until recently, and discern patterns and drivers of return without the bias or this effectively the single track of the hypothesis that it starts in a traditional quantitative process.
Now, I will say, and a lot of quants will hear me and say, “Oh, wait a minute. We use artificial intelligence.” They do use it in some respects. At least, you know, the more forward-looking they use natural language processing and other big data approaches. But the general principle still starts with a human being and a hypothesis there. The artificial intelligence just starts without their bias and then the potential bias of that looks at big data to find drivers of return. Does that make sense?
John: That does make sense. It brings up lots of questions. The good thing about a machine, potentially, is that it doesn’t feel emotionally attached to investment that’s losing money. So that’s a good thing. On the other hand, it’s taught by data and that data comes out of human activity and therefore human bias could creep into that.
Maybe let’s ask this question: What data does the AI access?
Robert: So, it runs the gamut, but it’s largely financial data that we are familiar with but at a depth and level that cannot be analyzed by a human being.
But there’s also things, such things as sort of unstructured data, less financial data that comes into play in AI investment process. So, things like web traffic to a particular website and what that might infer in terms of product popularity and the likes, you know, natural language, processing, and review of that.
So there’s a variety. It’s largely driven in investment management by traditional investment data we are familiar with, but done so at a scale that is unmatched. It literally creates trillions of potential combinations and derived data but also unstructured data can play into that. So, not that unfamiliar at the front end.
John: That brings up lots of questions, of course, because where the stock market moves is a function of so many different complex things, right? There’s fundamentals of a company. There’s fundamentals in an economy or a geography or whatever. There’s manipulation, there’s just people driving up prices or doing odd things. There’s speculation, right? There’s news and stuff happening as a result of that.
Robert: And there’s emotion.
John: And there’s emotion. I wonder what your AI would do in the case of a GameStop or an AMC where there’s almost zero rational reason behind what the market is doing, but yet it’s working in some sense.
Robert: Yeah. Well, so — probably oversimplifying it — on the financial aspects of it, it probably would not like it for reasons that you said, right? But artificial intelligence models are also looking at trading data and momentum and other factors, which could cause the model to like it in a much more shorter-term manner that to, you know, what you indicated, wouldn’t be really based on the traditional fundamentals and probably any derivation thereof, but it could still be based on momentum-driven factors, particularly trading momentum-driven factors.
Robert: So it is very much possible in that regard. But you’re right, there is a level of sort of clinician-ness, I guess, for lack of a better label — I don’t know, is that a word? — to the process and to the analysis of the financial data.
John: Yeah. Yeah, I’m guessing though it would look at the number of shorts on a stock. It would look at all that data and that will feed into it. And I’m guessing, I mean, this is machine learning, after all. I’m guessing it’s getting smarter all the time. It’s probably making predictions about what will happen with the stock, seeing if that prediction comes true. That reinforces certain things. If it doesn’t come true, then there’s an occasion for learning.
Is that accurate to say?
Robert: Yeah. That’s absolutely accurate. So typically, the way things work within artificial intelligence models within financial service, I’d say typical, because — and we’ll probably get to this later — it’s not that widespread yet, but the framework normally is that you have what is called a training window, whereby you are feeding, and that can go over several years, whereby you are feeding the model data and teaching it to interpret and link to excess returns and better performance.
And then you’re constantly retesting that to see if the model is learning with new data. That all happens in the training window and it’s rigorously, I’m oversimplifying, but it’s rigorously tested within that. And then you exit the training window and then it is making decisions going forward, always with data that is only available at the time.
So it’s very crucial as a rigorousness matter to avoid any look-forward bias. So we’re not ever asking it to make decisions with data that it had forward and then going backwards. So it’s never making any decisions with any data that it could not have known at the time. So, in simple terms, some data comes out well after its effective date. So, GDP data can be revised and revised and revised. It will not use that.
We work very, very hard to cleanse data at the point of consideration by removing biases, removing timing issues, removing survivorship bias issues in that, so that the model is only looking at data that was available at the moment.
John: Does it look at geopolitical data, like Russia declares war or a ‘special military operation’ on Ukraine?
Robert: Our model does not as of yet, because those are squishier things that are hard to… not all data that goes into AI models is purely quantitative as we talked about earlier. But there are things that there still needs to be research done in order to put it into a manner that a machine can contemplate and analyze. So there’s still more work to be done in that regard.
John: That said, I’m sure it’s seeing in real-time what’s going on with the stock market and adjusting as it goes because it sees all financial data. Correct?
Robert: Well, yeah, and I should have qualified. I’m glad you said that. It will see all of the financial data that, some of which is the result of that, some of which isn’t, but at the time, but it won’t be trying to interpret what’s Vladimir Putin’s next move and what would that mean for the markets and stuff. Not yet anyway. Not yet anyway.
John: Yes. Yes. So blow our minds here. How much better is this than a human? What kind of multiple on returns can we expect from this type of system?
Robert: Yeah, so I will answer your question, but as a precursor, I want to be clear that alpha remains very hard, and that doesn’t change with artificial intelligence processes. But what we are doing is trying to bring the human intuition at great scale, cost, and efficiency.
And we believe, I believe fundamentally, that artificial intelligence will substantively improve that process. It will not, if you sort of line it up against a fundamental manager, it will not beat them, you know, every time.
But we believe fundamentally it will do so more often than not, just because of its speed and analysis, and because of its efficiency, because of its scale. We have, the track records here, John, are relatively short as a statistical matter and statistician matter. We have I would’ve considered out of sample backtesting. So, not survivor, not just back-tested stuff but added sample testing, I should say, that goes in some cases back 10 years. Some statisticians would say, “That’s not enough, you need 20 to 30 years to prove unequivocally better,” but our records are far better than the average in-category in the majority of cases.
And our ETFs, the relatively short track record, you know, three years, almost, coming up on three years, have done quite well relative to their benchmarks, or I should say, well, relative to their benchmarks thus far.
So, you know, for us a really good achievement would be 100 basis points after cost, above the equivalent active manager in this space. And that’s our next goal in this area.
John: Gotcha. So, not a finance geek but 100 basis points is 1%. Is that correct?
Robert: 1%, yes, yes.
John: Okay. Very good. I thought I knew that, but I wanted to be sure about that. Okay. So 1% improvement after costs. It’s not blow me down, knock me over. It’s not 10% better. Is it gonna get there?
Robert: I don’t think it’ll ever get to 10%, John. The markets are still relatively efficient there. And I think that kind of outsized performance would probably only happen in very narrow asset classes where there’s a tremendous amount of inefficiency today. We are careful not to overpromise. We think it’s the better way to go. And in the long run, the market will come to realize that. But, again, as I said, it’s never gonna be meant to replace the human being in the process. There’s always going to be a role for them.
John: Right. So this is interesting, because if anybody ever developed … we’re gonna talk about your go-to-market, who you sell it to, who can use it, and all that stuff. But if anybody ever developed the one that’s 20% better or something like that, that’s probably an insane number. You’re in the space. I’m not in the space, so I don’t really know what’s a good number and what’s an insane number. But I’m guessing that’s insane in your thinking.
They would never sell it [laughing]. They would just push money into it and then run it.
Robert: Borrow every dollar they could and, yeah, do it themselves for their own worth. Yeah, absolutely. I mean, I’m still a believer in that some people should have a substantial portion of their assets in purely passive index investments. But I also believe that there, I think going forward, there’s going to be a lot of tailwinds to the adoption of artificial intelligence in investment management.
Robert: It’s been slow to be adopted quite frankly, far slower than many of the other areas we’re very familiar with, even if we don’t understand the technology really well. I mean, obviously, it’s fundamentally changed customer service, manufacturing, healthcare diagnostics, etc., etc. And it’s been slow to be adopted in asset management. I think it’s just a matter of time, and I think there’s a couple of tailwinds that we can talk about that I think is going to accelerate that.
John: So how does somebody use it? Can somebody come in on your website and sign up? Are you selling it to investment advisors? How does it work?
Robert: Yeah, so a couple different things. One is, and the most obvious way, to get for an individual to participate would be to invest in one of our exchange-traded funds traded on the New York Stock Exchange, and there is four of them in that regard. You are right to say that a big part of our focus in our business will be working through others that have those links to individual investors and institutions ultimately, because, you know, Qraft is small. It’s very expensive to reach the entire end investor market. We’re not a Fidelity. We don’t have a [inaudible] that will be, you know, we’re on Fidelity’s platform or ETFs. So we’ll be a combination of ETFs that investors can access, and we’ll be increasing the number of those over the next couple of years.
But also we’re going to provide — and already do — strategies to platforms, asset managers, like for their own use in their strategies, or potentially [co-working] opportunity with us.
John: Mm-hmm. A question occurs … why do you need the human? You’ve talked a couple times about how you’re pairing the scale and capacity of AI and machine learning and computers with the intuition of humans.
But theoretically, you have access to the entire history of a variety of stock exchanges, what’s gone on, what’s happened. You can feed the system other types of data as well, historical data, market data, climate data, you name it. And the machine theoretically ought to be able to learn to navigate financial markets and decide which stocks might be more valuable in the future.
Why is a human still involved?
Robert: Because, for a couple of reasons. The primary reason is we’re still sort of early days. Possibly in the long run what you say is true. I don’t think we know yet. But these machines still need monitoring in terms of new opportunities, new data sets that they could potentially utilize. And that requires some judgment on how well that’s being incorporated and being used over time. And we think that part probably never stops, in terms of the opportunity to learn further. So I would be thinking at least for the foreseeable future, that is always going to be there. But also there’s just this element of comfort and sort of oversight that we think is important and thoughtful.
John: Here’s all my money. I’m throwing it at the machine. Do what you will. Oops … made a mistake. It’s down to zero. [Laughing]
Robert: Exactly. So there’s this threat — I’m sure you’ve, you may not have heard it but we’ve already heard it and there’s an article today of, like, the concern that the machines will run amuck, so to speak — and I think that’s overblown greatly.
And I think it comes out of sort of just a fundamental misunderstanding of how artificial intelligence-driven processes are still the art form of human beings. And a failure in that, a failure of artificial intelligence or a failure to outperform, is still a failure of a human being.
What this is doing, ultimately, what artificial intelligence is doing, is trying to scale the human mind to plug into that intuition, but do so at a speed and scale that we can’t do as humans. That’s what it’s all about. And so, that oversight will always be there at least for, I think, as long as you and I are living. I don’t know how old you are, but for as long as I am living. I don’t think that goes away. And so this idea that the machines are gonna take over and run amuck I think is just, you know, foundationally a misunderstanding of how these things are put together.
John: Yeah. I guess it depends largely on whether we achieve general AI or not. But certainly the level of consciousness, if we want to put it that way, in AI right now is near zero in at least most machines. But maybe look into your crystal ball and look ahead five years to what you’re delivering, and where do you think the solution is then?
What do you think it’s able to do? What kind of returns in excess of average do you think you’ll be able to deliver?
Robert: Yeah, so, let me take the last part first and then go back on the other pieces. I don’t think there is going to be the case that the more time AI has that there’s sort of a linear multiplicative rise in investment returns relative to the averages. And that’s really what we’re always talking about, relative to the averages.
I think that as techniques get a little bit more precise and refined, there will probably be a rise in that. But I don’t think it ever becomes a, you know, beating the market by 4% or 5% each and every year, which would be in traditional investment terms is half a percentile type level performance. It’s very, very rare in any sustained way.
So I think this improves the process, refines the process. I think it drives cost out of the process. So, I think we will probably need less human beings, aggregately, in the investment management process as these things get adopted, just as quantitative techniques are less costly than before AI, than traditional, fundamental, you know, manage the stock at its most finite level.
So I think it improves but I don’t think it’s a multi, you know, 4% or 5% outperformance over time.
I think though, in terms of your earlier question, which I think is sort of where is it gonna be applied? How widely? What’s growth look like? I think is where you were going. I think it’s gonna be applied quite widely. I think the only limitations, John, are gonna be the education of the marketplace and people’s understanding of things. I think there is three or four really compelling tailwinds that are gonna drive wider consideration and use of AI in investment management. I think the reality of we’ve been in, you know, we’ve had 13 … putting aside the last couple of weeks, which has been pretty dreadful in the markets, we’ve got a 13-year bull market where it didn’t matter where you were invested, it just mattered that you were invested.
The concept of risk has gone out the window. We’re going to be in a much lower return environment almost by every person’s perspective over the next 10 years, meaning mid-single digits we’ll be lucky to get, just because we’ve basically borrowed forward over the last 10 years of sort of long-term returns because of the interest rate environment, etc.
And I think that’s gonna spur more interest in alpha and in just alpha strategies and alternative alpha strategies in a big way. And I also think the concept of risk management is gonna come back. I mean, I’ve had this conversation with many colleagues that I grew up in an industry where it wasn’t just about the return. It was also about the risk you were taking, but when you have a 13-year bull market with very little volatility, the concept of risk-managed returns has kind of gone out the window and…
John: Oh, risk management is hiring somebody from the SEC [laughing] and having them on your staff. Right?
Robert: Exactly. So, yeah, exactly. So it’s more business risk than investment. So I think that’s gonna come back.
And AI, I think, is uniquely positioned to be much more responsive to risk dimensions in the marketplace, as well as feed that need for sort of alternative alpha strategies.
I think, too, I think the rise of ESG, which is at its heart, is a data problem … largely, and AI is very effective in scaling and cleansing data. And I think we are very early days with the — I think ESG is going to fundamentally also change. So they’re the two big drivers I think over the next 10 years, and ESG is a little ahead in that regard.
John: And for those who may not know, ESG is Environment, Social, and Governance, which is basically about de-risking yourself from nasty businesses that are doing bad things.
Robert: Exactly. Exactly. And that is at its heart a data challenge for that to accelerate further, but I think it will unequivocally. And you know, that one of the other ones is the demographics and the aging of the population. You know, we’ve spent most of my, you know, life, the investment management industry has been about compelling people to accumulate for retirement.
Ten thousand people a day are turning 65 in this country. We are entering the de-accumulation period where people will now be drawing down their portfolios to sustain the living that they want in retirement. And that is an inherently very complicated problem.
And one, we, as an industry, don’t have a lot of familiarity with. And you say, “Well, why is it any [complicated], it’s still investing, right?” Well, it is, but with a very different dimensionality. It’s much more risky. Investment risk and the implications of it are much more substantial when you are in retirement, particularly in [inaudible] retirement.
If we have a major investment mistake, or a major market calamity when we’re accumulating, we have some options. We can double down on our savings. We can work longer. We have some options to fix things. If you’re a 75-year-old, you know, living off your portfolio for income and a major market calamity happens, you don’t have a lot of options. So it’s far riskier, and there’s different demands. There’s a different emotional dimension. You’re focused on income in a world where we have, as you well know, historically incredibly low-interest rates and the need to generate income for retirees.
Anyway, I could go on and on. I have real passion on this one. But I think artificial intelligence is uniquely positioned to deal with these new dimensions of consideration, and do so in a very, very responsive way in consideration of taxes, income, investment risk. I mean, it’s no elixir to a failure to save for retirement, to be clear. But I think the criticality of having a responsive process that can handle these new intricate complications is gonna lead to a lot more adoption of these processes.
John: Yeah. I think you’re probably right. And I think we’ll see a bit of an arms race in the fintech space around AI for investment and AI for assets and all those other things. And we’ll see, you know, somebody will have some stronger AI in this area or some better AI in that area, and we’ll see things go back and forth. And those who don’t adopt at all will be at a significant deficit of intelligence and data and insight, and probably suffer as a result.
Well, we are approaching some challenging times, and as you mentioned, you said 10,000 people a day are retiring. And that means a drawdown. That means sell pressure. That means a lot of stuff, right? And we are in unknown territory having had that bull market for forever and now so much geopolitical stress and strife, and just coming out of a pandemic, and supply chains still in a state of… [laughing] almost disarray. Very interesting times. I want to thank you for this time, Robert. Thank you for the explanation, and have a wonderful day.
Robert: Thank you so much for having me, John. It was a pleasure. Take care.
TechFirst is about smart matter … drones, AI, robots, and other cutting-edge tech
Made it all the way down here? Wow!
The TechFirst with John Koetsier podcast is about tech that is changing the world, including wearable tech, and innovators who are shaping the future. Guests include former Apple CEO John Scully. The head of Facebook gaming. Amazon’s head of robotics. GitHub’s CTO. Twitter’s chief information security officer, and much more. Scientists inventing smart contact lenses. Startup entrepreneurs. Google executives. Former Microsoft CTO Nathan Myhrvold. And much, much more.